# Historical Context — From Optical Dispersion to a General Principle *How K–K emerged from 1920s optics, acquired mathematical rigour in the 1930s, and generalised into the dispersion-relations programme of mid-century physics.* ## Nineteenth-century roots: dispersion without causality By the late nineteenth century, the wavelength dependence of the refractive index n(λ) of glasses and optical crystals was being fitted with empirical formulas. **Augustin-Louis Cauchy** (1830s) gave the first, `n(λ) = A + B/λ² + C/λ⁴`, useful in the visible but with no absorption terms. **Wolfgang Sellmeier** (1872) replaced it with a physically motivated sum of pole terms, ``` n²(λ) − 1 = Σ_k B_k λ² / (λ² − λ_k²) ``` whose poles at λ = λ_k were the absorption lines of the medium. Sellmeier already exhibited the qualitative link between absorption and dispersion — each pole produces a resonance feature — but the relation was only a modelling tool, not a theorem. **Hendrik Lorentz** placed this on physical foundations in the 1880s–90s with his electron theory. Each resonance became a damped harmonic oscillator, giving the complex susceptibility ``` χ(ω) = Σ_k (N e²/m ε₀) / (ω_k² − ω² − iγ_k ω) ``` and making explicit that the real (dispersive) and imaginary (absorptive) parts of χ arise from the same oscillator denominator. What was still missing was a general, **model-independent** statement: any causal medium must obey a relation of this form, not merely the Lorentz model. ## Kronig 1926, Kramers 1927 — what they actually did The two papers that gave the relations their joint name are cited far more often than read, and the textbook account of "what Kramers and Kronig did" usually credits them with the modern model-independent derivation. Bohren (2010) traced the original sources and showed the posthumous attribution is wrong on several counts. The accurate picture is narrower than the legend. - **Ralph Kronig** (1926, *J. Opt. Soc. Am.*) studied X-ray dispersion near absorption edges. Working from a Sellmeier-style sum over atomic oscillators with the polarizability *real* (his absorption coefficient was integrated over a narrow line), he derived an expression for `n(ω) − 1` in terms of the absorption coefficient — the **first known dispersion relation**. The paper contains *no complex variables*: no contour integration, no analytic continuation, no Cauchy integral formula. Only the n−1 relation is given, not its absorption-from- dispersion partner. And causality is not mentioned, even obliquely. - **Hans Kramers** (1927, *Atti Cong. Internaz. Fisici, Como*) presented a parallel derivation at the Como Volta-centenary congress. Working with the *atomic polarizability* `(ξ, η)` rather than with the bulk susceptibility, Kramers showed that its real and imaginary parts are Hilbert-transform pairs (without using the name). He ended by *speculating* that the same connection might hold for the real and imaginary parts of `ε − 1` in a medium denser than a gas — i.e., the dispersion relations on χ that we now call "Kramers–Kronig" were still a conjecture, not a result, in 1927. Causality and signal speed are not mentioned. Both derivations therefore appealed to a *specific atomic-gas model* — Sellmeier's `1/(Ω² − ω²)` denominator — rather than to the modern arguments from linearity, causality and analyticity. Both predate any rigorous treatment of convergence at infinity or of the subtractions needed when χ does not vanish at high frequency. Kronig and Kramers knew of each other's work; Kronig published first, which is how he ended up sharing credit for relations of which he had derived only one. ## Kronig 1942 — the first model-independent derivation The clean, model-independent derivation that physicists usually attribute to "Kramers and Kronig" was actually written by Kronig alone in 1942, in a Dutch-language journal (`Ned. Tijdschr. Natuurkd.` 9, 402–409). There he derived the full pair of relations on χ′(ω) and χ″(ω) without invoking any specific model: he recognised χ(ω) as the Fourier transform of a polarization response function, and invoked causality by *assuming the time-domain response vanishes for negative time*. From there the inverse cosine transform of χ′ equals the inverse sine transform of χ″, which is the K–K pair. He used no contour integrals, so his integrals are not written explicitly as Cauchy principal values, but the derivation is as compact as any modern textbook version. Even here, however, he did not mention signal-speed or Einstein causality — only "strict" causality, the statement that effect cannot precede cause. The mythology that Kramers used analytic continuation in the complex frequency plane to show that signals cannot travel faster than `c` (asserted by Toll in 1956 and repeated widely thereafter) is not in either Kramers's 1927 paper or Kronig's 1926 or 1942 papers. Kronig himself contributed to the historical confusion: in a 1936 paper with Gorter he cited his own 1926 paper as the source of the χ′-from-χ″ relation, even though that relation was first written down by Kramers in 1927 — and even then only as speculation. Both derivations predate modern rigour — the Cauchy integral formula is used informally, and convergence and subtraction questions (important when χ does not vanish sufficiently fast at infinity) are not spelled out. The cleaner statement awaited Titchmarsh. ## Mathematical context: Hilbert, Plemelj, Sokhotski The integral operator appearing in K–K, ``` (Hf)(x) = (1/π) P ∫_{−∞}^{∞} f(x′)/(x′ − x) dx′ ``` was already a classical object under the name **Hilbert transform**, introduced by **David Hilbert** around 1905 in his work on integral equations and the Riemann–Hilbert problem. Properties of its singular kernel had been studied earlier by **Julian Sokhotski** (1873, PhD dissertation, St Petersburg) and independently by **Josip Plemelj** (1908), who proved what is now the **Sokhotski–Plemelj formula** ``` 1/(x − x₀ ∓ iε) → P(1/(x − x₀)) ± iπ δ(x − x₀) as ε → 0⁺ ``` That identity is the key technical step behind K–K (Chapter 8): it says the boundary values of `1/(z − x₀)` from the two sides of the real axis differ by a delta function, which produces the pair of integral relations when causality is imposed. In 1937 **Edward Charles Titchmarsh**, in *Introduction to the Theory of Fourier Integrals*, proved what is now called **Titchmarsh's theorem**: for a function f ∈ L²(ℝ), the following three statements are equivalent. 1. f is the boundary value of a function holomorphic in the upper half-plane with bounded L² norm on horizontal lines (i.e., f lies in the Hardy space H²). 2. The Fourier transform of f is supported on [0, ∞) (one-sided in time, i.e., causal). 3. Re f and Im f are a Hilbert-transform pair — the K–K relations. Titchmarsh is the first water-tight derivation of K–K. It turns "the relations follow from causality" into a theorem of real-variable analysis. ## Dispersion relations as a general programme After the Second World War, the same structure was recognised far outside optics. The logic — causality (analyticity) plus an asymptotic bound (polynomial growth at infinity, handled by subtractions) gives a dispersion relation — became a workhorse of high-energy physics. - **Murray Gell-Mann, Marvin Goldberger, and Walter Thirring** (1954) wrote down the first rigorous dispersion relations for forward photon-nucleon scattering, turning K–K into a prediction for hadronic cross sections. - **Goldberger** (1955) and later work by **Bremermann, Oehme, Taylor** and others extended this to general scattering amplitudes and proved analyticity from microcausality in quantum field theory. - **Froissart, Martin, Mandelstam** (late 1950s–early 1960s) combined dispersion relations with unitarity to bound high-energy cross sections (the Froissart bound) and to develop the Mandelstam representation. Parallel developments occurred in: - **Nuclear physics**: optical-model analyses of nucleon–nucleus scattering use dispersion relations between real and imaginary parts of the optical potential. - **Acoustics**: causality in lossy acoustic media implies a K–K relation between phase velocity and attenuation (the O'Donnell relation, 1981). - **Electrical engineering**: Hendrik Bode's *Network Analysis and Feedback Amplifier Design* (1945) had already used the same gain–phase relation for minimum-phase transfer functions under the name **Bode relation**. - **Condensed-matter physics**: K–K analysis is the standard tool for extracting optical constants from reflectance spectra (the Kramers– Kronig-constrained reflectance method). ## Take-away K–K is best seen not as one result but as the prototype of a family. The causal-linear-system / analytic-transfer-function / dispersion-relation triangle reappears whenever: - time-domain response is one-sided (causal); - the response is linear and time-invariant; - the transfer function does not blow up at high frequency, or blows up only polynomially (allowing "subtracted" dispersion relations, Chapter 12). The next chapters lay down the complex-analytic machinery that makes this triangle into a theorem. ## See also - [01_introduction.md](01_introduction.md) - [03_complex_functions_analyticity.md](03_complex_functions_analyticity.md) - [08_principal_value.md](08_principal_value.md)